Covariate-Adjusted Difference in Proportions from Clinical Trials Using Logistic Regression and Weighted Risk Differences

Abstract

Risk differences and associated confidence intervals are frequently the basis for statistical testing in clinical trials. The analysis is often complicated by the presence of measured baseline covariates related to response that may be used to improve the precision of the treatment comparison by covariate adjustment in the statistical analysis. We use a clinical trial example and supporting simulations to show that logistic regression can be used to estimate the risk difference and compare its performance to common weighted-difference methods. We also examine when a useful improvement in precision can result from covariate adjustment, the use of a continuous rather than categorized covariate, and the consequences of including an unpredictive covariate.